RX 480 OpenCL Question

Message boards : Number crunching : RX 480 OpenCL Question
Message board moderation

To post messages, you must log in.

Previous · 1 · 2 · 3 · 4

AuthorMessage
Profile Darrell
Volunteer tester
Avatar

Send message
Joined: 14 Mar 03
Posts: 267
Credit: 1,418,681
RAC: 0
United States
Message 1867180 - Posted: 14 May 2017, 0:28:20 UTC - in response to Message 1867169.  
Last modified: 14 May 2017, 0:33:47 UTC

Arg, this drives me crazy:

https://1drv.ms/i/s!ArIvftV8roEagQvCbSdMgpMPz4ex

Host's daily average has gone from 380,000 per day to over 500,000 per day, in two days, and I don't know which change or combination of changes to the system have caused it.

It would appear Collatz Conjecture is paying stupid amounts of Credit for doing very little work.
30,000 Credits for less than 1,000 seconds work? That's not Credit inflation, that's Credit hyperinflation.

Let's see, if Seti paid like them, just one of my GTX 1070s would give almost 10,000 Credits per WU, instead of it's present 80-145.
Edit- unless they have a fixed pay per WU in which case i'd be getting 90,000 Credits for 1,000 seconds work.

What a joke.


I can not control what each project decides to pay, I try to counter Collatz by giving it the lowest resource share of the GPU projects. If a project has GPU and CPU apps, I just run the GPU apps. Resource share settings are PrimeGrid - 170%, Seti and Seti_beta - 160%, Einstein and Milkyway - 150%, MooWrapper - 145%, Collatz - 140%. The non-GPU projects are at 100%, except for Pirates which is stuck at 170% because its projects preferences update is broken. The mixture of tasks is up to the Boinc Scheduler. What I don't know is if this sudden surge is due to a change in the Boinc system settings { even making a change in RivaTuner settings can affect Boinc} or due to the having just Seti and Seti_beta tasks run for half a day while varying the command line settings.

The only good thing I can say about Collatz, is that while it is running it is only using 0.5% of a thread
ID: 1867180 · Report as offensive
Profile Wiggo
Avatar

Send message
Joined: 24 Jan 00
Posts: 34744
Credit: 261,360,520
RAC: 489
Australia
Message 1867184 - Posted: 14 May 2017, 0:47:41 UTC

Now those are very strange resource share numbers.

Basically you should be looking at the whole as 100% and then allocate shares as portions of that 100% (your numbers could be making it hard for the manager to work the proper shares). ;-)

I allocate 100% share to Seti while my backup projects get 0% so I only get work from those backup projects when I run out of Seti work and only when I allow them to.

Cheers.
ID: 1867184 · Report as offensive
Profile Darrell
Volunteer tester
Avatar

Send message
Joined: 14 Mar 03
Posts: 267
Credit: 1,418,681
RAC: 0
United States
Message 1867215 - Posted: 14 May 2017, 4:40:52 UTC - in response to Message 1867184.  

Now those are very strange resource share numbers.

Basically you should be looking at the whole as 100% and then allocate shares as portions of that 100% (your numbers could be making it hard for the manager to work the proper shares). ;-)

I allocate 100% share to Seti while my backup projects get 0% so I only get work from those backup projects when I run out of Seti work and only when I allow them to.

Cheers.


I look at it as two wholes, one for non-GPU tasks and one for GPU tasks. With three threads for non-GPU tasks and two for GPU tasks Boinc does a good job of rotating non-GPU projects on the CPU and GPU projects on the GPU. After major changes to the settings, or project outages, etc.it takes awhile for the system to stabilize, but when it does, Boinc does a good job rotating projects and downloading the next task before one finishes. Occasionally, it does get tripped up by a task that finishes before expected or project outage. Please note, that due to testing various settings and builds of Boinc, I normally run with my cache size set to zero, that way if anything goes wrong, I don't end up trashing a lot of tasks.
ID: 1867215 · Report as offensive
Profile Raistmer
Volunteer developer
Volunteer tester
Avatar

Send message
Joined: 16 Jun 01
Posts: 6325
Credit: 106,370,077
RAC: 121
Russia
Message 1867227 - Posted: 14 May 2017, 6:35:19 UTC - in response to Message 1866316.  

They take the stock apps and recompile them using all the compiler optimization settings possible to get the fastest code possible.

Do youspeak about OpenCL Windows binary for AMD GPUs?
Then who are "they" and what particular compilation options could be changed to speedup over current stock?
SETI apps news
We're not gonna fight them. We're gonna transcend them.
ID: 1867227 · Report as offensive
Profile betreger Project Donor
Avatar

Send message
Joined: 29 Jun 99
Posts: 11360
Credit: 29,581,041
RAC: 66
United States
Message 1867271 - Posted: 14 May 2017, 15:56:33 UTC - in response to Message 1867169.  

It would appear Collatz Conjecture is paying stupid amounts of Credit for doing very little work.

After crunching Collatz for a very short period of time I started feeling more stupid than usual.
ID: 1867271 · Report as offensive
Profile Darrell
Volunteer tester
Avatar

Send message
Joined: 14 Mar 03
Posts: 267
Credit: 1,418,681
RAC: 0
United States
Message 1867283 - Posted: 14 May 2017, 16:52:46 UTC - in response to Message 1867271.  
Last modified: 14 May 2017, 17:06:48 UTC

It would appear Collatz Conjecture is paying stupid amounts of Credit for doing very little work.

After crunching Collatz for a very short period of time I started feeling more stupid than usual.


Like I said what each project offers in credit, only they have control over. Another example is Einstein, when their tasks end, the amount of credit a tasks generates is less 1, usually 0.88 and they award successful tasks with 3645 credits. But what I was trying to get at when I first posted this about this, is that this host was generating a fairly steady 380,000 credits per day for all projects combined, and that suddenly after some change I made it has since started increasing drastically, it is currently over 686,000 per day and and does not show any signs of leveling out yet. And it just drives me crazy that I do not know what change or combination of changes has caused this.
ID: 1867283 · Report as offensive
Grant (SSSF)
Volunteer tester

Send message
Joined: 19 Aug 99
Posts: 13720
Credit: 208,696,464
RAC: 304
Australia
Message 1867395 - Posted: 15 May 2017, 4:34:00 UTC - in response to Message 1867283.  

this host was generating a fairly steady 380,000 credits per day for all projects combined, and that suddenly after some change I made it has since started increasing drastically, it is currently over 686,000 per day and and does not show any signs of leveling out yet. And it just drives me crazy that I do not know what change or combination of changes has caused this.

Pretty much all your Credit is from Collatz, and your RAC on that alone is around 480,000.
Grant
Darwin NT
ID: 1867395 · Report as offensive
Profile Darrell
Volunteer tester
Avatar

Send message
Joined: 14 Mar 03
Posts: 267
Credit: 1,418,681
RAC: 0
United States
Message 1867414 - Posted: 15 May 2017, 9:19:42 UTC - in response to Message 1867395.  
Last modified: 15 May 2017, 9:28:57 UTC

this host was generating a fairly steady 380,000 credits per day for all projects combined, and that suddenly after some change I made it has since started increasing drastically, it is currently over 686,000 per day and and does not show any signs of leveling out yet. And it just drives me crazy that I do not know what change or combination of changes has caused this.

Pretty much all your Credit is from Collatz, and your RAC on that alone is around 480,000.


That may be, but misses the point I was trying to make, so I'll show it in pictures:

Host average before change: https://1drv.ms/i/s!ArIvftV8roEagQ9xallrirTLILiW
Host average during change: https://1drv.ms/i/s!ArIvftV8roEagRF-tvylj0gPlG2k
Host average after another change: https://1drv.ms/i/s!ArIvftV8roEagRKn5cpANm2rAvYF

All of my GPU Projects:

Collatz: https://1drv.ms/i/s!ArIvftV8roEagR2-VP170n8acrfP <--- for some reason only showing last three days
Einstein: https://1drv.ms/i/s!ArIvftV8roEagRASX3gAwlvrscoV
Milkyway: https://1drv.ms/i/s!ArIvftV8roEagRPfWmZ57UqD9Jex
Moo!Wrapper: https://1drv.ms/i/s!ArIvftV8roEagRR3Zp-7IRhgyBlH
PrimeGrid: https://1drv.ms/i/s!ArIvftV8roEagRcB4BvQxx8t0UeO
Seti: https://1drv.ms/i/s!ArIvftV8roEagRr-SptYWytEoN-u
SetiBeta: https://1drv.ms/i/s!ArIvftV8roEagRkkUvIOqzIpM82W

Some of my CPU Projects:

Asteroids: https://1drv.ms/i/s!ArIvftV8roEagQ4-mqVWczLDueDY
Burp: https://1drv.ms/i/s!ArIvftV8roEagQ4-mqVWczLDueDY <--- I get a processor with more than two cores so I can run it now, and it runs out tasks the first day
NFS: https://1drv.ms/i/s!ArIvftV8roEagRX3iU51Q4DGJaji
Pogs: https://1drv.ms/i/s!ArIvftV8roEagRbRqc9fRsdc_k8t
Rosetta: https://1drv.ms/i/s!ArIvftV8roEagRiBqvKKI6b1cvTe
Universe: https://1drv.ms/i/s!ArIvftV8roEagRy2fosT5tgefQMy<-- project attached in April because they are working on getting a GPU version going.

Except for Collatz and Universe, all of the above pictures show data going back to March 16, 2017 when I upgraded the GPU from a Radeon HD5850 to a MSI Armor RX480 8GB OC. Also at this time is when Dr. Anderson released his first build of the Boinc system version 7.7.2. On April 24, 2017 I upgraded my processor from a 2 core Athon II X2 250 running at 3.0 GHz to a 6 core Phenom II X6 1100T running at 3.4GHz, unfortunately I only have a stock cooler, so I can not run it faster without risking CPU failure from overheating. My motherboard is a nine year old Biostar A780L3G with two 4GB sticks of DDR3-13333 memory made by Nano Technologies. I have a 1TB Sata hard drive and the system is powered by a Xion 1000W power supply. Note the motherboard only has one PCIEx16 2.0 slot to run the GPU. Somewhere around the 7th or 8th of May I built my current debug version of Boinc from the master branch of the repository, version 7.7.0. The HD5850 was a special GPU that had its core power locked, so I really could not overclock it. Now with new CPU and GPU I have been able to play with overclock settings. Unfortunately, I think I may pushed the GPU to far running a Moo!Wrapper task and fried a VRM which is why Boinc and OpenCL went from reporting 8192MB to 7536MB of GPU memory. All through these changes I have fought with GPU driver failures from TDRDelay timeouts, overclocking crashes, HWinfo64 crashing the GPU during startup detection, learning how to use MSI Afterburnner and AMD Overdrive and Wattman, learning how to fine tune Boinc Project apps, etc. In other words, over the last month and a half, I have had a lot of fun playing around with my computer. Running the HD5850 it sounded like I had a turbine engine in the room, with the MSI GPU, I can run its fans up to 100% and all I hear is a soft whoosh, it's only drawback is that it has driven the electric bill up by $40.
ID: 1867414 · Report as offensive
Grant (SSSF)
Volunteer tester

Send message
Joined: 19 Aug 99
Posts: 13720
Credit: 208,696,464
RAC: 304
Australia
Message 1867415 - Posted: 15 May 2017, 9:31:22 UTC - in response to Message 1867414.  

and that suddenly after some change I made it has since started increasing drastically

Correlation is not causation.

That may be, but misses the point I was trying to make, so I'll show it in pictures:
Collatz: https://1drv.ms/i/s!ArIvftV8roEagR2-VP170n8acrfP <--- for some reason only showing last three days

And you're missing the point i'm making, even though your Collatz graph shows it- your sudden boost in RAC is solely due to Collatz.
Grant
Darwin NT
ID: 1867415 · Report as offensive
Profile Darrell
Volunteer tester
Avatar

Send message
Joined: 14 Mar 03
Posts: 267
Credit: 1,418,681
RAC: 0
United States
Message 1867569 - Posted: 16 May 2017, 3:50:52 UTC - in response to Message 1867415.  
Last modified: 16 May 2017, 4:21:49 UTC

Ok Grant I'll concede that point, but here is another picture:

Boinc Credit Without Collatz: https://1drv.ms/i/s!ArIvftV8roEagR-3esX_xC-DU3wZ

Even though the graph only covers 30 days, by my calculation it is showing a 200% increase or 199.5753378% if you prefer exactness. Still not too shabby for a nine year old computer.

P.S. And this is with preference set to 51% and gpu_usage set to 0.5 and cpu_usage set to 0.17 resulting in a total of 5 tasks running, each getting 0.17% of the CPU or i.e 1 core or 1 thread (whichever you choose to call it).

P.S.S. Was offline for 9 hours today, I deleted something that took out the network, finally had to bite the bullet and do a system restore to get back online, also increased the core multiplier from 17 to 17.5 taking me from 3.4GHz to 3.5GHz, temps appear to be ok for now. The system is chewing thru Enigma tasks tonight, task times are about 1/4 what they were at 3.4, and are validating. Oops, my increasing clock speed, Enigma release a new version of their app while I was offline. Going from 1200 secs to 300secs is quite an improvement for an app.
ID: 1867569 · Report as offensive
Profile Darrell
Volunteer tester
Avatar

Send message
Joined: 14 Mar 03
Posts: 267
Credit: 1,418,681
RAC: 0
United States
Message 1868845 - Posted: 22 May 2017, 20:38:16 UTC - in response to Message 1867569.  
Last modified: 22 May 2017, 20:52:32 UTC

My apologies to Grant, Brent, Hal9000, and Wiggo. After more reviewing of the code and watching Boinc in process explorer, Boinc does automatically assign one core to each GPU tasks and that by using cpu_usage setting 0.17 it was indeed throttling the tasks. RAC daily without Collatz peaked at 159,434.5 on the 16th and currently is at 149,847.99 on the 22nd of May. But that reduction reflects a lot of time when Boinc was shutdown due to removing a lot of old crap off the computer, de-fragmenting the HDD, upgrading to the latest AMD GPU driver, system recovery when deleting something takes out the networking, etc. Once I get the system stable again, it should start to climb again.
... and still I fear, and still I dare not laugh at the Mad Man!

Queen - The Prophet's Song
ID: 1868845 · Report as offensive
Profile Darrell
Volunteer tester
Avatar

Send message
Joined: 14 Mar 03
Posts: 267
Credit: 1,418,681
RAC: 0
United States
Message 1872858 - Posted: 14 Jun 2017, 7:59:04 UTC

It's been awhile, but have been playing with the command line parameters and watching their effects. Finally figured out why Boinc went from reporting 8192MB of memory on the card to 7536MB back on April 19th. It was due to a bios setting involving the iGPU on the motherboard. Even though I have the iGPU disabled, you still have to set where shared memory is set. Choices are above 4gb or below 4gb. Setting it to above 4gb, Boinc and Opencl report 7536MB of memory on the card. Setting to below 4gb, Boinc and Opencl report 8192MB of memory on the card.
... and still I fear, and still I dare not laugh at the Mad Man!

Queen - The Prophet's Song
ID: 1872858 · Report as offensive
Profile HAL9000
Volunteer tester
Avatar

Send message
Joined: 11 Sep 99
Posts: 6534
Credit: 196,805,888
RAC: 57
United States
Message 1872983 - Posted: 14 Jun 2017, 22:45:18 UTC - in response to Message 1872858.  

It's been awhile, but have been playing with the command line parameters and watching their effects. Finally figured out why Boinc went from reporting 8192MB of memory on the card to 7536MB back on April 19th. It was due to a bios setting involving the iGPU on the motherboard. Even though I have the iGPU disabled, you still have to set where shared memory is set. Choices are above 4gb or below 4gb. Setting it to above 4gb, Boinc and Opencl report 7536MB of memory on the card. Setting to below 4gb, Boinc and Opencl report 8192MB of memory on the card.

Changing the iGPU memory settings sounded weird to me, but then I realized you have 8GB of system memory. So then it made sense.
There is something about video memory being mapped to system memory... or something along those lines.
SETI@home classic workunits: 93,865 CPU time: 863,447 hours
Join the [url=http://tinyurl.com/8y46zvu]BP6/VP6 User Group[
ID: 1872983 · Report as offensive
Profile Darrell
Volunteer tester
Avatar

Send message
Joined: 14 Mar 03
Posts: 267
Credit: 1,418,681
RAC: 0
United States
Message 1879086 - Posted: 19 Jul 2017, 19:58:47 UTC

Started a little test this morning with Seti_Beta, but first a little system recap:

GPU - MSI RX 480 Armor 8gb OC (factory clocked at 1291MHz core, 2000MHz memory)
CPU - AMD Phenom II X6 1100T BE (currently running at 3.4GHz)
Memory - 2x4gb Nanyo Technologies (DDR3 -1333 at 667MHz 9,9,9,24,33,1T)
Motherboard - Biostar A780L3G (9 years old, 1 PCIEx16 2.0 slot)
PS - Xion 1000W
OS - Windows 7 Professional 64bit
AMD driver 17.7.1 updated 7/12/2017
MSI Afterburner 4.4.0 Beta 12 updated 7/13/2017
HWinfo64 5.55-3205 Beta updated 7/13/2017
GPU Projects - Collatz, Einstein, Milkyway, MooWrapper, PrimeGrid, Seti, Seti_Beta, Xansons
Boinc 7.8.0 w/ 84% CPU setting giving 4 CPU tasks (each from a different project) and 1 GPU task, Caches set to 0

With the update of HWinfo, the GPU has been running the GPU projects at the factory OC settings. Here it is almost 6 days later and the GPU has not thrown one memory error (not normal). For the test I set all of the GPU projects to NNT except for Seti_Beta, then using Afterburner bumped the GPU to 1300MHz and then to 1310MHz one the first task. After that when each task got to 95% complete, I bumped the clock up another 10. Got the clock up to 1400MHz, still no memory errors, when I scrolled the System Explorer window to fast and caused the video driver to crash with a TDR Delay. The crash took out Radeon Settings and upon a restart of the system and then Boinc a NSF CPU task crashed. Once the Set_Beta was restarted at about 49% complete at the factory settings, bumped the core voltage from +6 to +24 with Afterburner and the clock back to 1400MHz. Going to leave it there for awhile to see how the task validations go and take a nap.

Here is a picture of the desktop: https://1drv.ms/i/s!ArIvftV8roEagXNkcsO5mMAvPYwS

This is a link to the task results (tasks after 16:00 UTC are from the speed test): https://setiweb.ssl.berkeley.edu/beta/results.php?userid=168&offset=0&show_names=0&state=0&appid=

2 tasks have validated so far....
... and still I fear, and still I dare not laugh at the Mad Man!

Queen - The Prophet's Song
ID: 1879086 · Report as offensive
Profile Darrell
Volunteer tester
Avatar

Send message
Joined: 14 Mar 03
Posts: 267
Credit: 1,418,681
RAC: 0
United States
Message 1879115 - Posted: 19 Jul 2017, 23:07:37 UTC

After a three hour nap (dang prostrate), 9 tasks have validated and the one that experienced the TDR crash is inconclusive. I wonder how high I can push it?
... and still I fear, and still I dare not laugh at the Mad Man!

Queen - The Prophet's Song
ID: 1879115 · Report as offensive
Al Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Avatar

Send message
Joined: 3 Apr 99
Posts: 1682
Credit: 477,343,364
RAC: 482
United States
Message 1879122 - Posted: 19 Jul 2017, 23:49:43 UTC - in response to Message 1872983.  

Changing the iGPU memory settings sounded weird to me, but then I realized you have 8GB of system memory. So then it made sense.
There is something about video memory being mapped to system memory... or something along those lines.
Yep, that is what I ran into with my 8-9 GPU machine that wouldn't recognize more than I think 6. I have the slots, but it doesn't play nicely so I am limited to 6, and I think it has something link 16 or 32 gig of RAM, but that doesn't have any bearing on this issue sadly.

ID: 1879122 · Report as offensive
Profile Darrell
Volunteer tester
Avatar

Send message
Joined: 14 Mar 03
Posts: 267
Credit: 1,418,681
RAC: 0
United States
Message 1879674 - Posted: 22 Jul 2017, 14:46:10 UTC

The increase in clock speed was no problem for Seti and Milkyway, but it was too fast for PrimeGrid. So I have set the GPU for 1300MHz on the core and 2050MHz on the memory. All projects run ok with this setting. After shutting down Boinc and running some Valley Benchmarks on the GPU, verified that HWinfo64 was working correctly and that AMD has done wonders with the 17.7.1 driver. Had the GPU up to 1450MHz core and 2100MHz memory with only a couple hundred memory errors, with previous versions of the driver it would have been throwing hundreds of thousands of memory errors.
... and still I fear, and still I dare not laugh at the Mad Man!

Queen - The Prophet's Song
ID: 1879674 · Report as offensive
Profile Darrell
Volunteer tester
Avatar

Send message
Joined: 14 Mar 03
Posts: 267
Credit: 1,418,681
RAC: 0
United States
Message 1880869 - Posted: 29 Jul 2017, 1:00:15 UTC

Does anyone know of any 64-bit OpenCl apps? From the AMD developer site, to use OpenCL 2.0 and higher extensions an app must be compiled with 64_bits. If a 64-bit app was created, then the app would see my RX480 has 8gigs of memory and I wonder how much more of that would be available for OpenCL than the current 3072MB of the 32-bit apps?
... and still I fear, and still I dare not laugh at the Mad Man!

Queen - The Prophet's Song
ID: 1880869 · Report as offensive
Profile Mike Special Project $75 donor
Volunteer tester
Avatar

Send message
Joined: 17 Feb 01
Posts: 34253
Credit: 79,922,639
RAC: 80
Germany
Message 1880919 - Posted: 29 Jul 2017, 8:06:25 UTC

All seti OpenCL apps are 32 bit only.
Raistmer tolld me that there would be no benefit in using 64 bit.


With each crime and every kindness we birth our future.
ID: 1880919 · Report as offensive
Previous · 1 · 2 · 3 · 4

Message boards : Number crunching : RX 480 OpenCL Question


 
©2024 University of California
 
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.